How LLMs Work: Build and Test an AI Assistant | Python Labs
Understand how LLMs work through visual explanations and Python labs for building, inspecting and testing an AI assistant. This 3.5-hour course combines lessons, worked examples and exercises; it is not a continuous live-coding recording.
LLMs from the Inside Out: Build, Inspect, and Test an AI Assistant
An original 14-chapter course from QuietLoom Media, narrated by Alex Mercer.
WHAT YOU WILL LEARN
• Tokenization, next-token probabilities and learning from prediction errors
• Attention and a small, inspectable transformer built with NumPy
• Running a trained local language model with llama.cpp and Qwen
• Chat templates, instruction tuning and preference-learning concepts
• Hallucinations, retrieval-augmented generation (RAG) and tool boundaries
• Evaluating an assistant, inspecting failures and improving a complete workflow
WHO THIS IS FOR
Beginners who know basic Python. Follow the explanations, pause at exercises and compare your reasoning with the worked results. Explanations and demonstrations preserve real outcomes, including failures.
CHAPTERS
00:00:00 A useful answer is not necessarily a true answer
00:12:29 Text, bytes and tokens
00:29:32 Turning scores into the next token
00:44:50 Learning from text, one mistake at a time
01:01:30 How attention connects the words
01:19:32 A transformer, assembled
01:35:15 Running a language model on your own machine
01:50:24 From text completion to an assistant
02:05:46 Feedback and preferences
02:20:54 Why a fluent answer can still fail
02:34:40 Give the assistant a handbook
02:50:56 Tools and the application boundary
03:06:56 Test the whole workflow
03:23:14 Put the whole course to work
JOIN THE DISCUSSION
After the attention lesson: what would you expect a token to attend to, and why?
After the capstone: which failure would you test next—retrieval, routing or the final answer?
If an explanation helps something click, like the video and share it with a friend learning AI. Subscribe to QuietLoom Media for more complete, practical AI courses.
BUILD YOUR FOUNDATIONS
Machine-learning lesson playlist: • Machine Learning for Beginners — Complete ...
Complete machine-learning course: • Machine Learning Full Course for Beginners...
TOOLS AND FURTHER READING
NumPy: https://numpy.org/doc/
llama.cpp: https://github.com/ggml-org/llama.cpp
Qwen local model: https://huggingface.co/Qwen/Qwen2.5-1...
Attention Is All You Need: https://arxiv.org/abs/1706.03762
InstructGPT: https://arxiv.org/abs/2203.02155
Direct Preference Optimization: https://arxiv.org/abs/2305.18290
Retrieval-Augmented Generation: https://arxiv.org/abs/2005.11401
ABOUT THE DEMONSTRATIONS
Original explanations, fictional library records and numerical teaching examples. The small teaching transformer starts with random parameters; it is distinct from the trained local model. Our transparent lexical retriever illustrates retrieval concepts. The capstone preserves a documented routing failure; its small development evaluation is not a production benchmark. This is not a reproduction of a commercial frontier model.
Production uses our established synthetic Alex Mercer narrator, generated illustrative artwork, and original code-driven educational visuals. English captions and chapter navigation are included. No third-party course footage or transcript is reused.
#LLM #ArtificialIntelligence #Python